RU-2861595-C1 - AUTOMATED SYSTEM FOR INTERPRETING RESULTS OF DIGITAL RADIOGRAPHIC INSPECTION - ARTIFICIAL ANALYSIS SYSTEM

RU2861595C1RU 2861595 C1RU2861595 C1RU 2861595C1RU-2861595-C1

Abstract

FIELD: image processing. SUBSTANCE: present invention relates to the field of image processing using artificial intelligence, namely to systems for automating the processes of recognising and determining geometric dimensions and coordinates of anomalies based on the results of digital radiographic inspection methods of girth welds of a gas transport system and assessing the compliance of the parameters of detected anomalies with the requirements of regulatory and technical documentation. An automated system for interpreting the results of digital radiographic inspection (artificial analysis system) comprises a module that receives data via a wired or wireless connection from a radiographic inspection device. The system also comprises an image quality assessment module, an image stitching and shifting module, and a measuring tape marker extraction module, which performs the function of identifying and extracting measuring tape markers on the image and simultaneously contains a detector and a recogniser. The system also comprises a defect recognition module and a weld quality assessment report generation module. EFFECT: increased level of process control and reduced risk of defects. 7 cl, 6 dwg

Inventors

Kostiukov Dmitrii Sergeevich
Pinigin Arsenii Dmitrievich
Sergiev Semen Aleksandrovich
Larin Aleksandr Viacheslavovich
Solonko Dmitrii Sergeevich
Sannikov Anton Aleksandrovich
Lapshin Nikita Sergeevich

Dates

Publication Date: 20260506
Application Date: 20250818

Claims (13)

1. An automated system for decoding the results of digital radiographic testing, containing:
- a module that receives data via wired or wireless communication from a radiographic inspection device that transmits data on welded joints, their markings and the obtained radiographic images in DCM format;
- a module for assessing the quality of an image, which is designed to automatically determine the compliance of an image with the requirements of current regulatory documentation and documentation standards for digital methods of radiographic control and its registration in the system;
- a module for stitching and shifting images, which performs the function of automatically creating panoramic images;
- a module for identifying marks of the measuring belt, which performs the function of determining and identifying marks of the measuring belt in the image and contains an algorithm for the automatic recognition and identification of a weld seam in the image, contains both a detector and a recognizer;
- a defect recognition module that automates the process of identifying and classifying defects in an image, and also links the coordinates of the detected defects to a measuring belt;
- a module for generating a weld quality report, which automatically generates a weld quality report based on data received from the defect recognition module.
2. The system according to paragraph 1, characterized in that the module receiving data via wired or wireless communication from the radiographic control device receives data in the form of DCM format files via a Wi-Fi connection with a mobile radiographic control device equipped with IRIS equipment.
3. The system according to paragraph 1, characterized in that it is equipped with a feedback module, which is intended for providing users with information about problems in processed images and storing this information for further analysis and use.
4. The system according to paragraph 1, characterized in that the module for assessing the quality of the image consists of a submodule for searching for a standard in the image and a submodule for refining the boundaries of the standard in the image.
5. The system according to claim 4, characterized in that the submodule for searching for references in a photograph is configured to raise all pixel values of the image to the power of 0.2, apply the equalization filter EqualizationFilter from the welbook_utils.image_processing library, perform two-fold image normalization with a normalization parameter of 0.05, apply the MedianBlur filter with a kernel size value equal to 3×3, pad the image to exclude situations where the reference is located on the boundary between frames, search for contours in the resulting image, evaluate the possibility of each contour being a reference, sort the contours according to their assessments, and select the contour with the lowest value.
6. The system according to claim 5, characterized in that the submodule for refining the boundaries of the standards in the image is configured to cut out a rectangular section of the image with the standard, determined by the minimum and maximum coordinates of the points of the standard, from the image obtained after twice performing image normalization with a normalization parameter of 0.05, applying the GaussianBlur filter with a kernel size of 1×3, binarizing the image cv2.threshold with a boundary of 0, searching for contours in the obtained image and adding them to the list of all contours, AdaptiveThreshold crop obtained after cutting out a rectangular section of the image with the standard with the parameters blockSize=21, constant=0.7, maxVal=255, applying the MedianBlur filter with a kernel size value equal to 3×3, searching for contours in the obtained image and adding them to the list of all contours, evaluating the possibility of each contour to be a standard, sorting the contours by their estimates and selecting the contour with the smallest value, determining the angle of inclination of the standard, rotating the crop of the rectangular section of the image by the obtained angle, determining the coordinates of the standard in the rotated image and cutting out the standard from the rotated image.
7. The system according to paragraph 1, characterized in that the detector used is a model based on fasterrcnn_resnet50_fpn from torchvision.models.detection, further trained on 100 epochs on labeled images; during inference, the SAHI library is additionally used, which divides the image into patches with a width of 2048 pixels with an intersection of 20% and runs them through the model, after which it produces global coordinates.

Description

AREA OF TECHNOLOGY The present invention relates to the field of image processing using artificial intelligence, namely to systems for automating the processes of recognizing and determining the geometric dimensions and coordinates of anomalies based on the results of digital methods of radiographic testing of annular welded joints of a gas transportation system and assessing the conformity of the parameters of the identified anomalies with the requirements of regulatory technical documentation. PRIOR ART With the current state of technology, the quality and speed of defect detection in images obtained through radiographic inspection depends on the qualifications of the flaw detection specialist, as recognition is performed manually and directly impacts the process of capital construction and repair of gas transmission system facilities. There is no system for generating reports and conclusions based on the results of assessing the compliance of identified defects with the requirements of regulatory documentation. A method for industrial radiographic inspection using a detector and a control device connected to the detector via a wireless or wired connection is known from the prior art (see RU2736074C1, published 11.11.2020). The essence of the invention consists in storing data on welded joints, their markings, and the obtained radiographic images in the memory of a control device connected to the detector via a wireless or wired connection. After turning on the X-ray radiation, radiographic images of the welded joints are obtained, and the markings for the radiographic images of the welded joints are entered automatically in digital form. A list of all welded joints subject to inspection during a work shift, containing at least the serial number of the welded joint and the full text of the marking, is entered in advance into the memory of the control device connected to the detector via a wireless or wired connection, which also stores the radiographic images of the welded joints obtained by the detector. One of the analogues of the claimed invention is a method for placing a digital marking on a radiographic image during industrial radiographic testing (see RU2822862C1, published 15.07.2024). The essence of the invention lies in the fact that in the control device, a method of image binarization is applied to the radiographic image of a welded joint obtained by the detector, as a result of which a binarized radiographic image is obtained, on which the area of placement of the digital marking is then determined, in which: it is taken into account that the area of placement of the digital marking must be determined against the background of the binarized radiographic image, must be the same size as the digital marking being placed, and must not intersect with other objects located on the radiographic image; the number of pixels in the binarized radiographic image, which are identified as the background of the image, is determined; The binarized radiographic image is processed by sequentially going through the pixels of the image, while calculating the number of pixels that are identified as the background of the image, and an area of the shape is selected on the binarized radiographic image that includes exclusively the pixels that are identified as the background of the image and corresponds to the size of the digital marking. A digital imaging method for radiographic testing of welded joints of thick-walled pipelines is also known (see CN112630237A, published 09.04.2021). The invention relates to a digital imaging method for radiographic testing of a welded joint of a thick-walled pipeline. The digital imaging method includes the following steps: S1, determining the diameter of the thick-walled pipeline to be tested and locating the inner center of the radiation source; S2, selecting and checking the technical indicators of the wire diameter and the distance between the pixels of the linear pixel quality indicator in accordance with the detection sensitivity requirements; S3, designing the selection of the focus size of the radiation source, designing the scanning scheme by adopting the optimal magnification principle, determining the optimal magnification and the error range, and determining the distance L1 from the radiation source to the surface of the root layer weld and the distance L2 from the surface of the root layer weld to the imaging screen in combination with the specific specification of the thick-walled pipeline; S4, initializing the digital beam radiography equipment and checking the balance and stability of the image; S5, exposure according to the preliminary order related to the radiographic scheme, placing a linear image quality indicator on the surface of the root pass weld, confirming the image quality and using a fixed interval as the basis for image length calibration; and S6, collecting associated images, evaluating defects and performing direct digital shooting for radiographic in